Fractional attention entropy and enhanced intrinsic time-scale decomposition for robust underwater target signal detection

• A novel Fractional Attentional Entropy measure incorporating a fractional-order attentional mechanism to robustly quantify signal complexity under challenging ocean noise. • An adaptive enhanced intrinsic decomposition technique improving upon traditional ITD for handling non-stationary and time-varying underwater signal components. • The hybrid FAE-EITD feature extraction framework, combined with a Decision Tree classifier, enables highly accurate (98.86%) identification of six ship types on real-world data, outperforming ITD-DispEn by 19.05%. • Demonstrates Robustness with Limited Data: The DT classifier shows excellent generalization capability (98.86% vs. PNN's 97.05%) using limited training samples, providing a practical solution for weak underwater target detection. Underwater target signal detection faces great challenges due to complex ocean noise and time-varying signal characteristics. In this paper, a novel hybrid framework combining fractional attentional entropy (FAE) and enhanced intrinsic time scale decomposition (EITD) is proposed to improve the robustness of feature extraction. FAE quantifies the signal complexity through fractional-order attentional mechanism while EITD adaptively decomposes the non-stationary components. The extracted features are fed into a decision tree (DT) classifier for six types of ship target recognition, which achieves an average accuracy of 98.86% on a real-world dataset. Comparative experiments show that the proposed method is 19.05% superior to the traditional intrinsic time scale decomposition (ITD) and dispersion entropy (DispEn). In addition, implementation with probabilistic neural network (PNN) shows that DT achieves excellent generalization with limited training samples (98.86% vs. 97.05% accuracy), and this research provides a theoretical basis for underwater weak target detection.